class Logger(object):
def __init__(self, vis_screen):
self.viz = VisdomPlotter(env_name=vis_screen)
self.hist_D = []
self.hist_G = []
self.hist_Dx = []
self.hist_DGx = []
def log_iteration_wgan(self, epoch, gen_iteration, d_loss, g_loss,
real_loss, fake_loss):
print(
"Epoch: %d, Gen_iteration: %d, d_loss= %f, g_loss= %f, real_loss= %f, fake_loss = %f"
% (epoch, gen_iteration, d_loss.data.cpu().mean(),
g_loss.data.cpu().mean(), real_loss, fake_loss))
self.hist_D.append(d_loss.data.cpu().mean())
self.hist_G.append(g_loss.data.cpu().mean())
def log_iteration_gan(self, epoch, d_loss, g_loss, real_score, fake_score):
print("Epoch: %d, d_loss= %f, g_loss= %f, D(X)= %f, D(G(X))= %f" %
(epoch, d_loss.data.cpu().mean(), g_loss.data.cpu().mean(),
real_score.data.cpu().mean(), fake_score.data.cpu().mean()))
self.hist_D.append(d_loss.data.cpu().mean())
self.hist_G.append(g_loss.data.cpu().mean())
self.hist_Dx.append(real_score.data.cpu().mean())
self.hist_DGx.append(fake_score.data.cpu().mean())
def plot_epoch(self, epoch):
self.viz.plot('Discriminator', 'train', epoch,
np.array(self.hist_D).mean())
self.viz.plot('Generator', 'train', epoch,
np.array(self.hist_G).mean())
self.hist_D = []
self.hist_G = []
def plot_epoch_w_scores(self, epoch):
self.viz.plot('Discriminator', 'train', epoch,
np.array(self.hist_D).mean())
self.viz.plot('Generator', 'train', epoch,
np.array(self.hist_G).mean())
self.viz.plot('D(X)', 'train', epoch, np.array(self.hist_Dx).mean())
self.viz.plot('D(G(X))', 'train', epoch,
np.array(self.hist_DGx).mean())
self.hist_D = []
self.hist_G = []
self.hist_Dx = []
self.hist_DGx = []
def draw(self, right_images, fake_images):
self.viz.draw('generated images',
fake_images.data.cpu().numpy()[:64] * 128 + 128)
self.viz.draw('real images',
right_images.data.cpu().numpy()[:64] * 128 + 128)
class Logger(object):
def __init__(self, vis_screen):
self.viz = VisdomPlotter(env_name=vis_screen)
self.hist_D = []
self.hist_G = []
self.hist_Dx = []
self.hist_DGx = []
def log_iteration_wgan(self, epoch, gen_iteration, d_loss, g_loss,
real_loss, fake_loss):
print(
"Epoch: %d, Gen_iteration: %d, d_loss= %f, g_loss= %f, real_loss= %f, fake_loss = %f"
% (epoch, gen_iteration, d_loss.data.cpu().mean(),
g_loss.data.cpu().mean(), real_loss, fake_loss))
self.hist_D.append(d_loss.data.cpu().mean())
self.hist_G.append(g_loss.data.cpu().mean())
def log_iteration_gan(self, epoch, d_loss, g_loss, real_score, fake_score):
print("Epoch: %d, d_loss= %f, g_loss= %f, D(X)= %f, D(G(X))= %f" %
(epoch, d_loss.data.cpu().mean(), g_loss.data.cpu().mean(),
real_score.data.cpu().mean(), fake_score.data.cpu().mean()))
self.hist_D.append(d_loss.data.cpu().mean())
self.hist_G.append(g_loss.data.cpu().mean())
self.hist_Dx.append(real_score.data.cpu().mean())
self.hist_DGx.append(fake_score.data.cpu().mean())
def plot_epoch(self, epoch):
# self.viz.plot('Discriminator', 'train', epoch, np.array(self.hist_D).mean())
# self.viz.plot('Generator', 'train', epoch, np.array(self.hist_G).mean())
# self.hist_D = []
# self.hist_G = []
hist = {}
hist['hist_D'] = self.hist_D
hist['hist_G'] = self.hist_G
np.save('training_hist.npy', hist)
def plot_epoch_w_scores(self, epoch):
self.viz.plot('Discriminator', 'train', epoch,
np.array(self.hist_D).mean())
self.viz.plot('Generator', 'train', epoch,
np.array(self.hist_G).mean())
self.viz.plot('D(X)', 'train', epoch, np.array(self.hist_Dx).mean())
self.viz.plot('D(G(X))', 'train', epoch,
np.array(self.hist_DGx).mean())
self.hist_D = []
self.hist_G = []
self.hist_Dx = []
self.hist_DGx = []
def draw(self, right_images, fake_images, path='fake_image.jpg'):
self.viz.draw('generated images',
fake_images.data.cpu().numpy()[:64] * 128 + 128)
self.viz.draw('real images',
right_images.data.cpu().numpy()[:64] * 128 + 128)
fake_images = fake_images.data.cpu().numpy()[:64] * 128 + 128
generated_images = fake_images
num = generated_images.shape[0]
height = int(np.math.sqrt(num))
width = int(np.math.ceil(float(num) / height))
shape = generated_images.shape[1:4]
"""
shape[0] - channels
shape[1] - height
shape[2] - width
"""
image = np.zeros((height * shape[1], width * shape[2], shape[0]),
dtype=generated_images.dtype)
for index, img in enumerate(generated_images):
img = img.transpose(1, 2, 0)
i = int(index / width)
j = index % width
image[i * shape[1]:(i + 1) * shape[1],
j * shape[2]:(j + 1) * shape[2], :] = img[:, :, :]
scipy.misc.imsave('Data/samples/' + path, image)
class Logger(object):
def __init__(self, vis_screen, save_path):
self.viz = VisdomPlotter(env_name=vis_screen)
self.hist_D = []
self.hist_G = []
self.hist_Dx = []
self.hist_DGx = []
self.logger = logging.getLogger('lossesLogger')
self.logFile = os.path.join('logs', save_path)
if not os.path.exists(self.logFile):
os.makedirs(self.logFile)
handler = logging.FileHandler(self.logFile + '/logFile.log')
handler.setLevel(logging.INFO)
self.logger.addHandler(hdlr=handler)
self.logger.setLevel(logging.INFO)
def log_iteration_wgan(self, epoch, gen_iteration, d_loss, g_loss,
real_loss, fake_loss):
#print("Epoch type: {} Gen iter type: {} dloss_type:{} gloss_type: {}, realloss_type: {}, fakeloss_type: {}".format(type(epoch), type(gen_iteration), type(d_loss), type(g_loss), type(real_loss), type(fake_loss)))
print(
"Epoch: %d, Gen_iteration: %d, d_loss= %f, g_loss= %f, real_loss= %f, fake_loss = %f"
% (epoch, gen_iteration, d_loss.data.cpu().mean(),
g_loss.data.cpu().mean(), real_loss.data.cpu().numpy(),
fake_loss.data.cpu().numpy()))
self.hist_D.append(d_loss.data.cpu().mean())
self.hist_G.append(g_loss.data.cpu().mean())
self.logger.info(
"Epoch: %d, Gen_iteration: %d, d_loss= %f, g_loss= %f, real_loss= %f, fake_loss = %f"
% (epoch, gen_iteration, d_loss.data.cpu().mean(),
g_loss.data.cpu().mean(), real_loss.data.cpu().numpy(),
fake_loss.data.cpu().numpy()))
def log_iteration_gan(self, epoch, d_loss, g_loss, real_score, fake_score):
print("Epoch: %d, d_loss= %f, g_loss= %f, D(X)= %f, D(G(X))= %f" %
(epoch, d_loss.data.cpu().mean(), g_loss.data.cpu().mean(),
real_score.data.cpu().mean(), fake_score.data.cpu().mean()))
self.hist_D.append(d_loss.data.cpu().mean())
self.hist_G.append(g_loss.data.cpu().mean())
self.hist_Dx.append(real_score.data.cpu().mean())
self.hist_DGx.append(fake_score.data.cpu().mean())
self.logger.info(
"Epoch: %d, d_loss= %f, g_loss= %f, D(X)= %f, D(G(X))= %f" %
(epoch, d_loss.data.cpu().mean(), g_loss.data.cpu().mean(),
real_score.data.cpu().mean(), fake_score.data.cpu().mean()))
def log_iteration_began(self, iteration, epoch, d_loss, g_loss, real_loss,
fake_loss, lr):
print(
"Step: %d, Epochs: %d, Loss D: %.9f, real_loss: %.9f, fake_loss: %.9f, Loss G: %.9f, lr:%.9f"
% (iteration, epoch, d_loss.data[0], real_loss.data[0],
fake_loss.data[0], g_loss.data[0], lr))
self.hist_D.append(d_loss.data.cpu().mean())
self.hist_G.append(g_loss.data.cpu().mean())
self.logger.info(
"Step: %d, Epochs: %d, Loss D: %.9f, real_loss: %.9f, fake_loss: %.9f, Loss G: %.9f, lr:%.9f"
% (iteration, epoch, d_loss.data[0], real_loss.data[0],
fake_loss.data[0], g_loss.data[0], lr))
def plot_epoch(self, epoch):
self.viz.plot('Discriminator', 'train', epoch,
np.array(self.hist_D).mean())
self.viz.plot('Generator', 'train', epoch,
np.array(self.hist_G).mean())
self.hist_D = []
self.hist_G = []
def plot_epoch_w_scores(self, epoch):
self.viz.plot('Discriminator', 'train', epoch,
np.array(self.hist_D).mean())
self.viz.plot('Generator', 'train', epoch,
np.array(self.hist_G).mean())
self.viz.plot('D(X)', 'train', epoch, np.array(self.hist_Dx).mean())
self.viz.plot('D(G(X))', 'train', epoch,
np.array(self.hist_DGx).mean())
self.hist_D = []
self.hist_G = []
self.hist_Dx = []
self.hist_DGx = []
def draw(self, right_images, fake_images):
self.viz.draw('generated images',
fake_images.data.cpu().numpy()[:64] * 128 + 128)
self.viz.draw('real images',
right_images.data.cpu().numpy()[:64] * 128 + 128)
class Logger(object):
def __init__(self, vis_screen):
self.viz = VisdomPlotter(env_name=vis_screen)
self.hist_D = []
self.hist_G = []
self.hist_Dx = []
self.hist_DGx = []
def log_iteration_wgan(self, epoch, gen_iteration, d_loss, g_loss,
real_loss, fake_loss):
print(
"Epoch: %d, Gen_iteration: %d, d_loss= %f, g_loss= %f, real_loss= %f, fake_loss = %f"
% (epoch, gen_iteration, d_loss.data.cpu().mean(),
g_loss.data.cpu().mean(), real_loss, fake_loss))
self.hist_D.append(d_loss.data.cpu().mean())
self.hist_G.append(g_loss.data.cpu().mean())
def log_iteration_gan(self, epoch, d_loss, g_loss, real_score, fake_score):
print("Epoch: %d, d_loss= %f, g_loss= %f, D(X)= %f, D(G(X))= %f" %
(epoch, d_loss.data.cpu().mean(), g_loss.data.cpu().mean(),
real_score.data.cpu().mean(), fake_score.data.cpu().mean()))
self.hist_D.append(d_loss.data.cpu().mean())
self.hist_G.append(g_loss.data.cpu().mean())
self.hist_Dx.append(real_score.data.cpu().mean())
self.hist_DGx.append(fake_score.data.cpu().mean())
def plot_epoch(self, epoch):
self.viz.plot('Discriminator', 'train', epoch,
np.array(self.hist_D).mean())
self.viz.plot('Generator', 'train', epoch,
np.array(self.hist_G).mean())
self.hist_D = []
self.hist_G = []
def plot_epoch_w_scores(self, epoch):
self.viz.plot('Discriminator', 'train', epoch,
np.array(self.hist_D).mean())
self.viz.plot('Generator', 'train', epoch,
np.array(self.hist_G).mean())
self.viz.plot('D(X)', 'train', epoch, np.array(self.hist_Dx).mean())
self.viz.plot('D(G(X))', 'train', epoch,
np.array(self.hist_DGx).mean())
self.hist_D = []
self.hist_G = []
self.hist_Dx = []
self.hist_DGx = []
def draw(self, right_images, fake_images):
right_images = self.rescale(right_images)
fake_images = self.rescale(fake_images)
self.viz.draw('generated images',
fake_images.data.cpu().numpy()[:16, :, :, :] * 255)
self.viz.draw('real images',
right_images.data.cpu().numpy()[:16, :, :, :] * 255)
def rescale(self, images):
images[:, 0, :, :] = images[:, 0, :, :] * 0.229 + 0.485
images[:, 1, :, :] = images[:, 1, :, :] * 0.224 + 0.456
images[:, 2, :, :] = images[:, 2, :, :] * 0.225 + 0.406
return images